Australia's eSafety named AI companions a child-safety risk

What happened

On 18 February 2025 Australia's eSafety Commissioner published an online safety advisory titled 'AI chatbots and companions, risks to children and young people'. It states that by early 2025 more than 100 AI companion apps existed, naming Character.AI, Replika and Talkie among them, many free and marketed to look attractive to young users, and often lacking age-restriction mechanisms. The advisory links high-frequency, problematic use of poorly safeguarded services to self-harm, citing the suicide of a 14-year-old boy in the United States without naming him.

What the documents show

The advisory lists specific risk categories: exposure to dangerous concepts such as inaccurate advice on self-harm, suicide or eating disorders; dependency and social withdrawal from overstimulated reward pathways; unhealthy attitudes to relationships from interactions that carry no real consequences; heightened grooming risk from normalised sexualised conversation; bullying; and financial exploitation through subscription upsells. Separately, eSafety's industry codes page, as retrieved on 16 September 2026, records that the second phase of Online Safety Codes, covering age-restricted material such as pornography, split into two tranches: three codes registered 27 June 2025 and in force from 27 December 2025, and six more, covering app-distribution platforms among other service types, registered 9 September 2025 and in force from 9 March 2026.

The mechanism

eSafety's advisory is guidance and public education, not a binding instrument in itself; the enforceable mechanism is the Online Safety Act's codes and standards, which the advisory says eSafety will use, alongside Basic Online Safety Expectations, to hold industry accountable. The app-distribution- platform code category is broad enough to reach a stand-alone companion app distributed through an app store, which matters editorially because many companion apps are not social media platforms in the conventional sense and might otherwise sit outside a narrower code.

What it leaves open

The advisory does not itself carry penalties; it is eSafety's public explanation of risk, paired with a pointer to codes that were still being phased in through most of 2025 and into 2026. Whether a specific companion app is judged to fall under the age-restricted material codes depends on service classification questions the codes page does not resolve for every product type. This is an editorial reading: a chat-only companion app without pornographic content sits in a greyer part of the age-restricted framework than an app that plainly hosts explicit material.

• Does the app enforce any age-verification step beyond a self-declared birth date?
• Which of eSafety's Online Safety Codes, if any, would classify this specific app as a regulated service?
• Does the product's design include the addictive engagement patterns the advisory flags, such as unprompted emotional questions?

Read together, the advisory and the codes show a regulator naming a risk in plain public language well before the enforceable code machinery for that risk had fully come into force.